HiHiAllen/Imaginarium-Dataset

Imaginarium是一个新颖的视觉引导3D布局生成系统，旨在解决生成逻辑连贯且视觉上吸引人的定制场景布局的挑战。该系统使用图像生成模型将提示表示扩展为图像，并通过微调使其与高质量资产库对齐。然后，开发一个强大的图像解析模块，基于视觉语义和几何信息恢复场景的3D布局，并使用场景图优化场景布局以确保逻辑连贯。数据集包含3D场景和资产，如Blend源文件、RGB渲染、实例分割、边界框、深度图以及元信息（如标题、场景图和对象姿势）。

Imaginarium is a novel vision-guided 3D layout generation system that addresses the challenges of generating logically coherent and visually appealing customized scene layouts. The system employs an image generation model to expand prompt representations into images, fine-tuning it to align with a high-quality asset library. It then develops a robust image parsing module to recover the 3D layout of scenes based on visual semantics and geometric information, optimizing the scene layout using scene graphs to ensure logical coherence. The dataset includes 3D scenes and assets such as Blend source files, RGB renders, instance segmentation, bounding boxes, depth maps, and meta-info (e.g., captions, scene graphs, object poses).